Análise preditiva do preço da cesta básica na cidade de Natal-RN utilizando redes neurais e séries temporais

Abstract

This study aims to reflect on the complexity of time series prediction techniques, focusing specifically on the price of the basic food basket in the city of Natal-RN. In an effort to contribute to the improvement and precision of forecasts, we compare the performance of two distinct methods: the traditional ARIMA model and two newer approaches based on neural networks, specifically Feedforward Neural Networks (FNN) and Multilayer Perceptrons (MLP). The research involved the collection of data related to the price of the basic food basket in Natal-RN, followed by the execution and training of the ARIMA, FNN, and MLP models. To compare the results obtained by each of these methods, performance metrics such as Mean Absolute Error (MAE) and Mean Squared Error (MSE) were considered to assess the robustness of each approach. The results revealed, in addition to the relative effectiveness of these techniques for predicting basic food baskets, insights into accuracy under different scenarios. In conclusion, the study emphasizes the relevance of considering approaches whose methods are currently being developed, particularly in situations involving dynamic contexts.